A study of SPEs and the consequences of radiation exposure

(1)

IN

DEGREE PROJECT VEHICLE ENGINEERING, SECOND CYCLE, 30 CREDITS

STOCKHOLM SWEDEN 2017,

A study of SPEs and the consequences of radiation exposure

SIGNE BJÖRNHOLDT BÖLL

KTH ROYAL INSTITUTE OF TECHNOLOGY

(2)

TRITA AVE 2017:15 ISSN 1651-7660

(3)

A study of SPEs and the

consequenses of radiation exposure

Author:

Signe Björnholdt Böll signebb@kth.se

Royal Institute of Technology (KTH)

Supervisors:

Christer Fuglesang Oscar Larsson

February 22, 2017

c

Signe Björnholdt Böll, 2017

(4)

(5)

Till mamma och pappa.

(6)

(7)

Abstract

With thorough literature studies as well as simulations, a way to minimize the exposure to radiation that astronauts are at risk of encountering during a solar proton event is sought. The understanding of where these particles come from, as well as the random nature of solar particle events is of importance in order to predict their occurrence. Different models used for predicting solar particle events based on a Poisson possibility distribution are presented, as well as real-time forecasts which give a warning of an approaching event. Although the models used for real-time forecasts have a high accuracy rate, the average warning time is only approximately one hour. The downside with the predicted possible occurrence is that this only gives a statistical probability of events that could possibly occur. For the real-time forecasts the downside is that with an average warning time of only one hour, they do not give a lot of time for seeking shelter during the onset of an event. With simulations it is shown that the best way to minimize the radiation dose obtained by astronauts is to use different materials of shielding. It is also shown that a lower shielding thickness when encountering SPEs, for example when in a space suit, is useful as long as the total amount of time spent in this suit during the duration of a mission is planned thoroughly in order to stay below the radiation dose limits. If an astronaut would be caught in an event with the same magnitude and intensity as the solar particle event of August 1972, it is shown that the astronaut only has nine minutes to seek shelter before exceeding the radiation dose limits and thereby risking radiation induced sickness.

Sammanfattning

Med djupgående litteraturstudier, samt simuleringar, undersöks möjligheten att minimera den strålnings- dos som astronauter riskerar att utsättas för under en protonstorm från solen. Förståelse för varifrån dessa laddade partiklar kommer, och även deras slumpmässiga beteende, är av vikt för att kunna förutspå dessa fenomen. Diverse modeller som används för att förutsäga hur dessa solstormar sannolikt kommer inträffa i framtiden, tillsammans med modeller som presenterar en prognos i realtid, diskuteras. Även om de tillgängliga realtidsmodellerna har en hög noggrannhet, så är ändå varningstiden inför en solstorm i medel omkring endast en timme. Detta ger väldigt lite tid för att söka skydd om en protonstorm skulle inträffa. Nackdelen med de förutsägelser som bygger på sannolikhet är att det endast ger en indikation på vad för slags protonstormar som skulle kunna inträffa. Med simuleringar visas att det bästa sättet att minimera strålningsdosen är genom att använda olika material samt tjocklek för att skärma av och skydda astronauten mot inkommande strålning. Simuleringarna visar även att ett tunnare skydd, som tex av en rymddräkt, kan vara användbart så länge den totala tiden som spenderas i strålningsmiljö planeras varsamt och delas upp mellan olika skyddsmaterial. Detta för att ej överskrida gränsen för accepterad strålningsdos. Slutligen visas även att en astronaut som befinner sig i en protonstorm med samma intensitet som den protonstorm som inträffade i augusti 1972 endast har 9 minuter på sig att söka skydd innan den accepterade dosen av strålning är överskriden och risken för akut strålningssjuka ökar.

(8)

(9)

1. Introduction

Since 1843 it has been known that the Sun, being a dynamic source of energy with constant dramatic events taking place on its surface, follows a periodic cycle. The cycle follows the variation in number of sun spots on the Earth’s surface, where the occurrence of big solar flares and coronal mass ejections intensifies as the solar cycle maximum approaches [Schwabe, 1844]. As seen in table 1, the solar cycle is on average 11 years long and stretches from minimum to minimum.

Table 1: List of solar cycles tracked since 1755, with a mean value of 11 years [Kane, 2002].

Cycle Started Finished Duration max. monthly

(years) sunspot number Solar cycle 1 1755 August 1766 March 11.3 86.5 (June 1761) Solar cycle 2 1766 March 1775 August 9.0 115.8 (Sep 1769 Solar cycle 3 1775 August 1784 June 9.3 158.5 (May 1778) Solar cycle 4 1784 June 1798 June 13.7 141.2 (Feb 1788) Solar cycle 5 1798 June 1810 September 12.6 49.2 (Feb 1805) Solar cycle 6 1810 September 1823 December 12.4 48.7 (May 1816) Solar cycle 7 1823 December 1833 October 10.5 71.5 (Nov 1829) Solar cycle 8 1833 October 1843 September 9.8 146.9 (Mar 1837) Solar cycle 9 1843 September 1855 March 12.4 131.9 (Feb 1848) Solar cycle 10 1855 March 1867 February 11.3 98.0 (Feb 1860) Solar cycle 11 1867 February 1878 September 11.8 140.3 (Aug 1870) Solar cycle 12 1878 September 1890 June 11.3 74.6 (Dec 1883) Solar cycle 13 1890 June 1902 September 11.9 87.9 (Jan 1894) Solar cycle 14 1902 September 1913 December 11.5 64.2 (Feb 1906) Solar cycle 15 1913 December 1923 May 10.0 105.4 (Aug 1917) Solar cycle 16 1923 May 1933 September 10.1 78.1 (Apr 1928) Solar cycle 17 1933 September 1944 January 10.4 119.2 (Apr 1937) Solar cycle 18 1944 January 1954 February 10.2 151.8 (May 1947) Solar cycle 19 1954 February 1964 October 10.5 201.3 (Mar 1958) Solar cycle 20 1964 October 1976 May 11.7 110.6 (Nov 1968) Solar cycle 21 1976 May 1986 March 10.3 164.5 (Dec 1979) Solar cycle 22 1986 March 1996 June 9.7 158.5 (Jul 1989) Solar cycle 23 1996 June 2008 January 11.7 120.8 (Mar 2000) Solar cycle 24 2008 January Still ongoing - 81.9 (Apr 2014)

Mean 11.1 114.1

In combination with these solar flares and coronal mass ejections, an acceleration of particles can sometimes occur with particles being thrown out into space from the Sun’s surface. Figure 1 shows daily sunspot number compared to solar activity and the space environment between 1983-2012, where it can be seen that the protons measured by GOES satellites follow the periodical cycle of the average number of sunspots, known as the solar cycle.

(12)

Figure 1: An overview of the space environment between 1983-2012 showing the daily Sunspot number compared to solar activity. From top to bottom the graphs illustrates the daily sunspot number, cosmic rays, measured X-rays, measured protons and the GOES magnetometer. [Wilkinson, 2012]

These solar particle events, or SPEs, can be hazardous to both astronauts and spacecrafts in space.

As travelling to Mars has become a soon possible reality, astronauts will spend more time in space making the risk of exposure to SPEs much higher. Therefore it is important to know more about SPEs, how and when they occur and also how to protect astronauts if they are exposed to them.

The total radiation an astronaut is exposed to in space consists of both radiation from SPEs and protons, as well as the long-term radiation from the galactic cosmic ray. In low earth orbit there is also a radiation contribution from trapped particles in Earth’s magnetic field. This thesis will only focus on the radiation originating from SPEs built up by protons, where the aim of this thesis is to investigate the possibility to predict solar particle events in order to minimize the risk of exposure and to maximize the time of warning. Another aim of this thesis is to investigate the use of shielding as a way to minimize radiation on the human body.

Literature studies of historical events, the possibility to predict future events, radiation impact on the human body as well as simulations of the radiation level for different materials of shielding will be performed in order to do so. The knowledge attained from the literature studies is discussed in chapter 2, and 4 while the simulations and results based on the literature studies will be discussed and presented in chapter 3 and 5.

2. Understanding solar particle events

The Sun, being a highly effective particle accelerator, can accelerate and launch particles out into space sometimes reaching relativistic speeds. This makes it possible to cover the distance between the Sun and Earth in a matter of minutes. Though humans on Earth are somewhat protected from these events thanks to Earths magnetic field and atmosphere, it can still cause severe damage.

Penetrating neutrons caused by high-energy proton induced fragmentation could be a threat to passengers and crew of high-altitude aircrafts on trans-polar routes [Reames, 2013]. It is also estimated that an SPE of extreme magnitude could cause damages to the U.S. alone at a cost of

(13)

several trillion dollars [Wei et al., 2013].

Since some claim that the possibilities of traveling further into space to distant planets like Mars are soon within our reach, where there is no magnetic field offering protection, astronauts and space craft face an even higher risk of being exposed. Protons of 150MeV can penetrate 7.4 cm of aluminium, equivalent to 15.5 cm of water or human flesh [Reames, 2013]. But already at energy levels of∼30 MeV will protons begin to penetrate spacesuits and the outer layer of a space craft. Therefore, an understanding of SPEs and their impact on the human body is crucial for space craft design as well as for planning missions in space. While it is possible to predict future SPEs to some extent, these predictions are either based on probability theory or they are real-time forecasts but with only a short time of warning before they occur. Where they come from, their levels of energy and when they will occur are questions that need to be considered, as well as the understanding of why they are dangerous for the human body in order to make space missions as safe as possible.

2.1. The theory of origin

The origin of a solar proton events is believed to be found at the heliosphere. Here, the protons are emitted from the Sun and accelerated to very high energies after which they follow the interplanetary magnetic field lines into space, becoming a solar particle event, as shown in figure 2. The SPE can be either magnetically well connected or magnetically poorly connected. The characteristics of the event depend on how the protons are accelerated. A proton accelerated during a solar flare or coronal mass ejection event (CME) is often magnetically well connected with the originating solar flare or CME, and an SPE consisting of protons accelerated in this matter is regarded as magnetically well connected as well. The SPE consisting of this type of protons is characterized by the protons rapidly rising intensities, thus making the intensity of the SPE rise rapidly. The proton can also be accelerated in interplanetary space by shock waves driven out from the Sun by CMEs. These protons are magnetically poorly connected with the originating solar event that accelerated the protons, making the SPE built up by this kinds of protons regarded as magnetically poorly connected. This type of event is characterized by a slow increase in proton intensities where the protons reach a maximum intensity after crossing through the shock wave, to the region where the magnetic fields connect to the shock nose from behind.

This is the modern theory of the origin of solar particle event, but this theory succeeds a more than 40 year old misunderstanding of the origin of solar energetic particle events [Gosling, 1993].

In 1946, Scott Forbush presented an article with a theory explaining an unusual increase in intensity during large magnetic storms occurring in February and March 1942 [Forbush, 1946].

This article would later be known as the first time that high-energy particles from the Sun had been observed at the Earth’s surface. These early observed SPEs are what we today refer to as ground-level events (GLE). At this time, the solar flare was already a know phenomenon, as it had been observed for the first time in 1859 [Carrington, 1860]. The knowledge of solar flares in combination with the observations made by Forbush led to the theory that the energetic particles in SPEs where accelerated at or directly above the site of a solar flare and that the energy required for the acceleration was derived from the strong magnetic fields in the flaring region.

(14)

Figure 2: The interplanetary magnetic field, originating in the Sun. [Nelson, 2016]

Even though this theory was not all wrong, there are more phenomenons occurring at the solar surface than just the solar flare. For low frequency radio waves, there are different types of radio bursts emissions that all originate in the same layer of the solar atmosphere [White, 2007]. The most interesting ones for studying solar activity and space weather phenomena are called type II, type III and type IV. During a solar flare, both X-ray and γ-rays are emitted. As the soft X-ray (i.e.

X-rays of lower energy) reaches its peak a type II radio burst typically occurs. This type of burst is identified by slowly decreasing to lower frequency levels with time. Type III bursts on the other hand occur on the onset of the solar flare and then drift rapidly to lower frequency levels with time. The Type IV bursts are not as common as the type II and type III bursts and are therefore somewhat less studied. But it is known that they are typically observed as the solar flare starts to decay. In 1963, a contradicting theory was presented, explaining that there where two types of physical mechanisms occurring at the Sun which were both responsible for accelerating energetic particles into space [Wild et al., 1963]. Wild et. al. based their theory on radio observations and proposed that there were two different types of SPEs correlating to radio bursts emitted from the Sun. The theory suggested that electrons were accelerated to produce type III bursts, while protons were accelerated at shock waves seen as type II bursts. Although the authors later on revised this theory by explaining that none of the radio emissions are produced directly by protons, it still identified two different acceleration mechanisms which where dominated by electrons and protons, respectively.

(15)

Coronal mass ejections were not yet discovered at this time, making it impossible to link one of the two acceleration mechanisms to these types of event. But as CMEs where later discovered in 1971 [Tousey, 1973], Kahler et. al. was able to prove a correlation between CMEs and SPEs a few years later [Kahler et al., 1978]. In 1999, Donald Reames published an article explaining particle acceleration at the Sun, explaining the origin for the different types of solar particles. This theory is now the fundament of today’s general understanding of the origin of SPEs [Reames, 1999].

As Reames explained in [Reames, 1999], and also later in [Reames, 2013], there are two types of solar particle events, originating from two different sources of acceleration, as shown in figure 3. Gradual SPEs are efficient, large, intense and extensive. Impulsive SPEs are the opposite; in relation to space these events are small, weak and compact, but also numerous. But how are the solar particles accelerated and what sets their acceleration apart?

Figure 3: The propagation paths of impulsive SPEs following the spiral field lines (A), and the gradual SPEs originating from CME shocks (B). [Nelson, 2016]

2.1.1 Type III bursts and impulsive SPEs

In 1965, Van Allen and Krimigis observed electrons with energies∼40KeV originating from the Sun at the same time as type III radio bursts [Van Allen and Krimigis, 1965]. This would be the first discovery of electrons that can generate type III radio bursts. Later on, Lin studied events with

>22keV electrons and found what he called "pure" electron events (i.e. events with no measurable levels of protons or other ions) that were also connected to solar flares and type III bursts [Lin,

(16)

1974]. The discovery of these pure events would later lead to the understanding of³He-rich events and their association with the 10-100 keV electrons that produce type III bursts [Reames et al., 1985]. This observation made it possible to identify³He-rich impulsive flares associated with these events. Eventually, the theory that all solar energetic particles accelerated in flares were

3He-rich, even while the energetic particles in SPEs seen in space were not (later to be known as gradual SPEs), was widely accepted [Reames, 2013]. Impulsive SPEs have also been found to have an abundance of³He/⁴He>0.1, which is much higher than the abundance ratio in the corona or solar wind of∼ ⁵×10⁻⁴, where some events have been found to have an abundance ratio as high as³He/⁴He>10. In large eruptive flares, a magnetic reconnection can occur beneath the CME. If this happens among closed magnetic field lines, the reconnected field lines of the loop must also be closed. This means that any³He-rich SPEs are locked in, and will eventually scatter into the loss cone and later the low corona. One theory is that since the particles have no way of escaping, the accumulated energy of the particles heats up the flare, helping to increase its magnitude. Impulsive SPEs are also commonly associated with jets, where the reconnection takes place between open and closed field lines, accelerating electrons and ions into open field lines.

These electrons stream outwards, producing Langmuir waves and type III bursts. Even though the type III bursts are generated during electron transport, and not by a physical acceleration mechanism, they are still used as the signature of an impulsive SPE.

2.1.2 Type II bursts and gradual SPEs

In 1984, Kahler et. al. found a correlation between large SPEs (i.e. gradual events) and fast, wide CMEs of 96%. In the same report it was also shown that ³He-rich events are strongly associated with impulsive flares, and that the physics of ion acceleration can explain the difference in abundance ratios between flares and shocks.

Many years later, in 1999, the evidence showing that the the CME drives the shock that produces type II bursts was presented [Cliver et al., 1999], ultimately connecting the type II burst with the SPEs. Looking at the surface of a solar shock wave, the electrons drift in the Vshock×Belectric field. In a limited region of the shock surface where the magnetic field lies near the plane of the shock, the electrons are accelerated, which generates Langmuir waves. These Langmuir waves interact to produce an electromagnetic radiation observed as a type II burst.

Gradual SPEs are not really gradual at all, but in comparison to impulsive events which only last a few hours, gradual events do have a relatively long duration of several days. This long duration is partly due to the fact that the particles are continuously accelerated by the shock [Reames, 2013]. It has been shown that that the solar longitude of the nose of the shock can provide information about the solar longitude of the maximum intensity of the particles near the time of shock passage. Although there is a correlation between the intensity of solar energetic particles and a CME and the shock speed, this only describes a general behavior of a very complex situation. In the report [Kahler, 2001] it was shown that the peak intensity of the solar particles at different energies correlates with the CME speeds. It also led to the realization that only the fastest 1-2% of CMEs are able to accelerate particles of SPEs. But what is the physics behind this acceleration? Reames explains in the report [Reames, 2013] that there are four distinct phases which modify the intensities, abundances and energy spectra of the event. These phases are the onset, the plateau, the shock peak and the reservoir, with the first and last phase applicable to both impulsive and gradual events.

Protons that are streaming out from a shock wave near the Sun generate Alfvén waves that scatter the particles that follow. If the intensity of protons at the source is increased it causes an

(17)

amplification of these waves, which in turn increases the scattering. When the proton meets the Alfvén wave it does so with a pitch angle. For a small angle, the result is an amplification of the wave. When the pitch angle of the proton is∼90^◦, the energy level of the proton increases when resonating with the amplified wave. It is believed that this phenomena, when the low-energy protons create a scattering environment, is what makes it possible for protons to reach higher levels of energy and can explain their rapid enhancement in energy. The intensity of protons will continue to increase until reaching an upper limit, creating an intensity plateau. It has been observed that intensities of protons seen early in large gradual events are bounded at an upper limit of∼200-400 protons/(cm2 sr s MeV), making it the streaming limit of the protons.

The acceleration of particles from CMEs is very complex, but the modern understanding is that it could be explained by the possibility that behind the CME and shock a reservoir is formed.

When the particles flow out they encounter a CME, and begin to fill the magnetic flux tubes that contain them. In other words, these particles begin filling the reservoir. This reservoir is capable of trapping high-energy particles over a large area for an extended period of time. The filling of the reservoir is slow, but leakage of particles is minimal, so that the intensity in the reservoir persists as the overall intensity declines.

A reservoir formed by one event contributes with the seed population needed for the acceleration by the shock of a second event. An SPE originating from a shock wave can sample particles from a previous event in the reservoir, building up its intensity. It is believed that the particle acceleration in an event is enhanced by a previous SPE and in a series of large SPEs, with a reservoir filled with particles from previous events, intensities can build up reaching the intensity levels of so called "super-events". Before the discovery of CMEs, gradual events were believed to originate from solar flares and their gradual behavior was explained by a slow diffusion of the particles in interplanetary space due to scattering. Now, with a broader understanding, it is believed that the slow decline in intensity is not a consequence of interplanetary scattering, but occurs because of the slow expansion of the volume of the reservoir.

3. Prediction of events

It is very hard to predict SPEs due to the random nature of their occurence. Yet it is important to be able to do so when planning a future space mission or EVA. The only way to predict SPEs that is available today is by using statistical models. These models can be compared to the statistical models of predicting the weather; the shorter time span the more accurate the prediction will be.

These models use the accumulated knowledge about previous events in such a way so that one can derive a probabilistic model which gives an indication for when large solar particle events could occur. There are several models available where most models use the set of data available at the time they where produced, meaning that for each new and upgraded model there is a smaller error in the predictions. For the simulations in this report, ESA’s Space Environment Information System SPENVIS [Heynderickx et al., 2000] has been used. Using SPENVIS, it is possible to predict long-term solar particle fluences, explained more thorougly in section 3.1, which are predicted for the total mission duration. In SPENVIS there is also an option of using models for predicting short-term solar particle fluxes which are predicted for single events, as discussed in section 3.2.

(18)

3.1. Long-term prediction

For predicting SPEs for a long duration of time when using SPENVIS, there are essentially three different models available: the King model, the JPL model and the ESP model. An update based on recent years of solar fluence data has later been implemented, called the Rosenqvist et. al model.

3.1.1 The King model

The King model [King, 1974] is based on data collected from the major solar proton events during solar cycle 20, between the years 1964-1976, and then uses this data in order to calculate the probability of events encountered for space missions performed during the next coming solar cycle. King argues that since the occurences, fluences and spectra of previous solar events have a quasi-random nature, it is possible to derive a statistical model of solar proton fluences relevant to space missions of given durations and orbital characteristics flown at a certain time in the future. The most significant aspect of this report is the distinction between ordinary events (OR), and what is called anomalously large events (AL) such as the event of 1972. This distinction is somewhat understandable since the event of 1972 stands for a major part of the total of proton fluences (69-84% for different levels of energy) during the seven year active phase of the solar cycle.

The fluences of ordinary events are for the King model described by a log normal distribution function, i.e. a probability distribution in which the logarithm of the intensities is distributed normally, and the probability of their occurences are derived based on this assumption. King continues with the most important conclusion in this report, the finding that "if the probability of having no AL-event is less than some specified confidence level for a given mission duration, then it is permissable to neglect the occurence of ordinary events in determining mission fluence".

There are 25 events included in the set of data used in this analysis, where the flux of protons observed at Earth and with an energy level higher than 10 MeV exceeds 2.5×10⁷/cm². These 25 events include all 20 periods in which the flux energy level > 30 MeV exceeds 5.0×10⁶/cm²and all 19 periods in which the flux energy level > 60 MeV exceeds 1.0×10⁶/cm².

King also mentions the fact that if all the events of cycle 19 (between 1954-1964) and 20 were used a statistically more significant data base would be obtained, but continues to say that the event-occurence rate and the generally larger event fluences of cycle 19 (compared to cycle 20) demonstrate that cycle 19 and 20 were not statistically similar, and that a Poisson distribution would not be suitable to derive from this larger set of data. King then continues to say that is is probable that cycle 21 (between 1976-1986) will be more similar to cycle 20 than to cycle 19, and therefor the analysis made for this model is restricted to the set of data obtained from cycle 20 for obtaining predictions for cycle 21.

In King’s report, it is shown that the probability P of encountering solar protons, for specific levels of kinetic energies E, during a space mission that exceeds the level of F is given by

P(>F, E; τ) =

∑

∞ n=1

p(n, τ, N, T) ×Q(>F, E; n) (1)

where the probability Q, with probability density q(F, E), that the logarithm of the combined fluence due to an observed amount of events that n will exceed F,

Q(<F, E; n) = Z _∞

−∞q(x, E)[>log(10^F−10^x), E; n−1]dx. (2)

(19)

It is also shown that the probability p of observing exactly n events in a future interval of a certain duration τ depending on the amount of events were observed in a past interval is a binomial distribution given as

p(n, τ; N, T) = (n+N)!

n!N! × (τ/T)ⁿ

[1+ (τ/N)]⁽1+n+N) ⁽³⁾

Where F is the base-10 logarithm of the proton fluence, τ is the duration of the space mission, n is the number of observed events in a future interval of duration τ, N are the number of events observed in a past interval of duration T. All of these equations are to be used for when no AL-event is expected to occur. King concludes the report by saying that for a given confidence level and mission duration, if an AL-event is expected, the OR-events can be neglected, which is also shown.

Table 2: The table shows how the dominant proton type varies with inclination and altitude.

Type of orbit low altitude mid altitude high altitude low-inclination orbit trapped protons trapped protons solar protons high-inclination orbit solar protons trapped protons solar protons

In 1975, King & Stassinopoulos released an updated version of King’s previous model [King and Stassinopoulos, 1975]. This new version included the ratios of solar to trapped proton fluences.

This update takes the following into consideration: the orbit altitude and inclination, mission duration, proton energy threshold as well as what level of risk the mission planner is willing to take that the actually enclountered solar proton fluence will exceed the predicted fluence levels. The report shows that trapped protons are dominant for low inclination orbits at low- or mid-altitude and for high-inclination orbits at a mid-altitude orbit. Solar protons on the other hand are dominant for high-inclination orbits with low altitude, as well as for low- and high-inclination orbits with a high altitude. This is also shown in table 2.

3.1.2 The JPL model

For planning space missions and to predict proton fluences, King’s model was the one being used during the 1980’s. In 1988, Feynman et. al presented an idea of how the King model could be upgraded and applicable for energies >10 MeV [Feynman et al., 1988]. The authors of the report claim that there are two big misinterpretations being made for the previous model. In King’s report, an assumption is used stating that the number of great proton flares during a solar cycle was a function of the cycle’s maximum sunspot number (which were regarded as established fact at the time when King’s model was presented). King also came to the conclusion that the sunspot maximum for cycle 21 would resemble or be smaller than cycle 20. With these assumptions it was reasonable to use the set of data from cycle 20 when predicting the fluence for cycle 21, but it has since then been found invalid. Therefor the authors suggest reviewing the data and to produce a new model.

At the time of the report by Feynman et al., the data set used for analysis was much bigger and consisted of both observations made from the Earth’s surface between 1956 and 1963 and observations made from spacecrafts between 1963 and 1985. With this new set of data, the previous assumption that there where two different kind of events made by King would be proven wrong.

(20)

Although basing the new analysis of the data on the previous work made by King, it was no longer assumed to be possible to distinguish ordinary events from the previously named anomalously large events as King regarded the event of 1972 to be. Instead, this new study was to be carried out by regarding all of the events to be ordinary. King also made a clear distinction between the maximum and minimum phases of the sunspot cycle, while Feynman et al. claim that these are not clearly defined. Instead they use a different approach and prove that the hazardous period for major proton events seems to extend from two years prior to a sunspot maximum till four years after the maximum. They regard this to be of importance when planning future space missions since you have to take into account the actual launch date of the mission.

The probability functions in this report are practically the same as the ones used for the King model. With fpbeing the proton fluence of an event, it can be written as fp=₁₀^Fwhich means that if fpis distributed lognormally then F is distributed normally.

The probability P that during a mission length τ the fluence level will exceed fpif

P(>F, τ) =

∑

∞ n=1

p(n, wτ)Q(F, n) (4)

where Q(F, n)is the probability that the sum of all fluences due to n events will exceed 10^F and p(n, wτ)is the probability of n events occuring during mission lenght τ if an average of w events occured per year. p(n, wτ)can be written as

p(n, wτ) =e^wτ(wτ)ⁿ

n! (5)

i.e. a Poisson distribution. The fluences attained with this new models are about twice as high as the ones attained by King for energies >10 MeV, which implicates that the King model does not suffice when calculating expected fluences during space missions. The authors finish by saying that since this model is not applicable for energies >30 MeV a suggestion is to use this model for energies >10 MeV and then extrapolate the results using the 1972 event as a model. Later in 1990, the authors present a revised version of the model which is also applicable for energies >30 MeV [Feynman et al., 1990].

In 1993, an updated model for predicting interplanetary fluences of protons is presented [Feynman et al., 1993] and is now called the JPL model. Now the data set covers a period from 1963 to 1991, which increases the accuracy of the previous model. The biggest difference is that this model is applicable for proton fluence energy levels >1, >4, >30 and >60 MeV. The new data confirms King’s opinion that the distribution of sizes of the proton events is such that the predicted fluence will be mostly influenced by a small number of very big events. In other words, it is very important to estimate the probability of large events occuring in order to get an accurate prediction of proton fluences. A comparison with the JPL model from 1985 (refered to as JPL-85) and the model from 1991 (refered to as JPL-91) shows a similar prediction, but the JPL-91 is shown to be more reliable due to the fact that the data set used for analysis has a much larger time span. The equations for deriving the possibilities of different scenarios are basically the same as for the previous models, and the function φ_total for estimating long-term solar system body exposure has been added, given by

φ_total= Z

F(p)f(p)dp (6)

where F(p)is fluence for a single solar cycle at a given probability level and f(p)is the probability distribution function, with probability p. The updated model is recommended for the case of single-event effects, total dose and dose rate effects as well as solar panel degradation.

(21)

In 2002, the authors behind the original JPL model present a review of the current JPL-91 model in the light of yet a new set of data including the previous years [Feynman et al., 2002]. According to the authors, their conclusion is that the JPL model is still sufficient and the previous assumptions made are justified. Finally, the name is changed to The JPL proton fluence model.

3.1.3 The ESP model

The model for Emission of Solar Protons, or the ESP model, was born in 1999, as a way of predicting the fluences of a worst case solar proton event [Xapsos et al., 1999]. In 2000 a revised version of the model that had the ability to predict the total fluence during the whole duration of a space mission was presented [Xapsos et al., 2000]. Arguments as to why this model is necessary, is that a spacecraft designer needs to take into account both the long-term cumulative proton fluence as well as the worst case events that are expected to occur, when designing a spacecraft.

Before the ESP model was available, it was often assumed that the worst case event a spacecraft could encounter would be of the same magnitude as the event of 1972. During this time, the recommendation from NASA was to consider the possible worst case event as a combination of the two events from February 1956 and Augusti 1972. Instead of using these methods, a reliable probabilistic model would be helpful to the designer as he or she could predict the worst case fluences as a function of confidence level and mission duration. It would also be helpful to set a desired upper limit for the maximum fluence. These two features are the essential characteristics of the ESP model.

As seen in the previous sections, the King model as well as the JPL model both rely on assuming the fluence distributions to be lognormal. For the ESP model, the authors have taken a different approach. Since there is a lack of data for larger events, they instead use the maximum entropy principle. The maximum entropy principle is a mathematical procedure for generating a probability distribution for incomplete data, and would therefor be better suited in this situation.

For deriving this model, solar proton event data from the last three complete solar cycles were used (i.e. solar cycle 20-22), and only events with a minimum fluence φminwere taken into account.

As in previous models, only events during the active years where analysed.

The first step is to find a function describing the initial distribution of fluence for the solar proton event. The distribution’s entropy, S, is defined as

S= − Z

p(M)ln[p(M)]dM (7)

were p(M)is the probability density of the random variable M. M is set as the logarithm of the event fluence, φ. The author uses the Lagrange multiplier technique, where λ is a Lagrange multiplier and b=λln(10). The solution p(M)which maximizes S is needed to solve the equation for N:

N=Ntot[^φ

−b−φ_max^−b φ_min^−b −φ_max^−b

] (8)

where N is the number of events per solar active year with a fluence greater than or equal to φ, Ntotis the total number of events per solar active year with a fluence greater than or equal to the minimum event fluence φ_minand φmaxis the maximum event fluence.

This can then be used to derive the equation for the worst case solar proton event fluences. If FT is the cumulative probability, then the worst case distribution for T solar active years is given by

FT(M) =exp(−NtotT[1−P(M)]) (9)

(22)

where F_T can be considered equal to the desired confidence level, and P(M)is the corresponding cumulative distribution to the probability density p(M). With M=log φ, P(M)can be written as a function dependent only of the event fluence levels:

P(φ) = ^φ

−b min−φ^−b φ^−b_min−φ^−b_max

. (10)

A good thing about this model is that the maximum event fluence φmax can be set as an upper desired limit when designing, but it is discussed in the article that due to the limited amount of data this feature has some uncertanties. As there was limited data for solar proton energies >100 MeV, and the results from using this model are valid between energy levels of > 1MeV to >100MeV, it is suggested to extrapolate between the energy levels >100 MeV and > 300 MeV. This in order to make the model applicable for this energy spectra as well.

For predicting the amount of accumulated fluence over a whole mission it was previously common to use either the SOLPRO model based on The King model or the JPL model. Both of these models have limitations due to the small amount of data that they are based upon. Another limitation is the proton range. The SOLPRO model covers the energy levels from >1 MeV up to 100 MeV while the JPL model covers energy levels between >1 MeV and > 60 MeV compared to the ESP model with from >1 MeV up to >300 MeV.

For accumulated fluences, the ESP model uses the maximum entropy theory for deriving these predictions as well. Though for this case the maximum entropy theory shows that the best choice of a probability distribution for the cumulative fluence Φ is a lognormal distribution and can therefor be written as:

FCU M= ¹ σ

√2Π Z _Φ

0⁺

1

Φ⁰exp{− ¹

2σ²[ln(Φ⁰) −µ]²}dΦ⁰ (11) where FCU M is the confidence level for observing a total proton fluenceΦ over a time period of T active years, and the lognormal parameters σ and µ are dependant on T.

The article concludes with comparing the ESP model to the JPL model with the conclusion that the two models are somewhat alike. There is a small difference where the ESP model gives slighty higher fluences than the JPL model. This is explained with the fact that the set of data used for ESP model has added the values from solar cycle 20-22, compared to the JPL model which only used cycle 20-21 and some parts of cycle 22. Since cycle 22 is considered to be a rather active cycle, this could explain the difference in the results according to the authors.

3.1.4 The Rosenqvist et al. model

The Rosenqvist model is viewed as a tool for analysing the previous models and determining their accuracy [Rosenqvist et al., 2005]. The authors themselves call it a "Toolkit for updating interplanterary proton-cumulated fluence models". The name gives a good indication of its intended use. When studying what impact the size of data sets and the derived values from this data have on the outcome when using the models for predicting proton fluence, it was shown that the values µ, σ and w used in the JPL-model led to an underestimated value of the fluence. The Rosenqvist model has since then been implemented in SPENVIS to be used as a combination with the JPL-91 model for predicting fluences.

(23)

Table 3: Input data for a trajectory representing the ISS trajectory around Earth during 1 year.

Input Value

Mission start January 1st, 1977

Mission end January 1st, 1978

Apogee 409.50 km

Perigee 400.20 km

Inclination 51.64^◦

Right ascension of ascending node 242.45^◦

Argument of perigee 27.71^◦

3.1.5 Comparing the models

Since the King-model is first and foremost designed for predicting solar proton fluences during the 21 solar cycle, and especially for the years between 1977-1983, the simulations for comparing the different models for predicting fluences are set to be derived for this time span. The simulated mission is for the ISS, which is in orbit around the Earth with an altitude of∼400 km. The exact values chosen for the simulation are given in table 3. The mission duration is set to 1 year and is starting on January 1st 1977. These values are to ensure that the mission is during the solar cycle maximum. The inclination of ISS is 51.64^◦[Peat, 2016], and the same inclination is used for this simulation, this is also the case for the other values. The apogee is set to 409.5 km and the perigee is set to 400.2 km, with the argument of perigree set to 27.71^◦.

Figure 4 shows the fluence predicted by the different models. The predicted fluence is plotted against the corresponding kinetic energy level using Matlab. As it can be seen in the figure, the King model shows a much smaller value for the predicted fluence compared to the other models.

The JPL model also predicts higher fluence levels than the King model, which was mentioned in the article [Feynman et al., 1990]. In the articles about the ESP model, it is mentioned that the ESP model gives a slightly higher fluence than the JPL model. In figure 4 this is shown to be true for the ESP model of total fluence for energy levels <55MeV, but not for the ESP model of a worst case scenario. The models give roughly the same result up to an energy level of 100 MeV, where the King model suddenly has a more rapidly declining function compared to the JPL and the ESP model. This could be explained by the fact that when the King model was produced, there where limited data for solar proton energies >100 MeV, and the King model is therefore not as reliable as the other models for this interval.

(24)

Figure 4: The predicted solar proton fluence level vs proton kinetic energy at the ISS during 1 year.

The same comparison is made for an orbit with an altitude of 400 000 km, which could represent the proton fluence one would be exposed to on the Moon. The exact values for this scenario are given in table 4. The apogee and perigee is set to 405,504 km and 363,396 km, respectively [TimeAndDate, 2016]. Looking at the predictions made by the different models for this new scenario, as shown in Figure 5, the results are similar with the exception of higher values of the proton fluence for higher altitude. This can be explained by the fact that ISS is shielded by the Earth’s magnetic field, while some of the orbital path of the Moon goes further than the reach of the magnetic field and thereby leaving the Moon more exposed to radiation.

Table 4: Input data for a trajectory representing the Moon’s trajectory during 1 year.

Input Value

Mission start January 1st, 1977 Mission end January 1st, 1978

Apogee 405504 km

Perigee 363396 km

(25)

Figure 5: The predicted solar proton fluence level vs proton kinetic energy at the Moon during 1 year.

3.2. Short-term prediction

It is essential to not only focus on the long-term predicted events when planning a mission, you also have to take the short-term predictions into consideration if any EVA is planned to occur.

In SPENVIS, the option of looking into the short-term solar particle fluxes is available to predict instantaneous fluxes. Different models for short-term predictions can be chosen in SPENVIS, these models are discussed and compared in the following subsections.

3.2.1 The CREME models

For short-term predictions, one option in SPENVIS is to use CREME. CREME was first presented in 1981, and is considered the first computer code to explain the complexity of the space ionizing- radiation environment [Tylka et al., 1997]. The name CREME is an abbreviation for Cosmic Ray Effects on Micro-Electronics, and is a program developed to study the effects of an ionizing radiation environment [Falzetta et al., 2007]. CREME includes models of the galactic cosmic rays (GCRs), anomalous cosmic rays (ACRs), solar particle events as well a geomagnetic transmission calculations. These models are then analyzed in order to create a numerical model of the ionization radiation and to build a particle flux at an orbit. CREME can then describe both the external

(26)

enviroment with no shielding, as well as the internal environment which is shielded. The first model of CREME is called CREME86, and the later updated model, which was presented in 1997, is called CREME96.

When using CREME86, the weather index M has to be set [Adams Jr, 1986]. With this index, it is possible to vary which sources and solar conditions is being used when calculating the cosmic ray flux near Earth. Presented in table 5, it is shown that there are four different type of events to choose from: Ordinary flare flux, 10% worst-case flare flux, Aug. 4, 1972 and the composite worst-case flare flux. The weather index is then divided into different scenarios, mean composition and worst-case composition, for each specific event.

Table 5: Weather index, M Input Value

5 peak ordinary flare flux and mean composition 6 peak ordinary flare flux and worst-case composition 7 peak 10% worst-case flare flux and mean composition 8 peak 10% worst-case flare flux and worst-case composition 9 peak Aug. 4, 1972, flare flux and mean composition 10 peak Aug. 4, 1972, flare flux and worst-case composition 11 peak composite worst-case flare flux and mean composition 12 peak composite worst-case flare flux and worst-case composition

For CREME96, the option of choosing different values of the weather index M is no longer available. Instead there are three options for this setting: worst week scenario, worst day scenario and peak 5-minute-average fluxes.

3.2.2 The ESP model

Another model available in SPENVIS for short-term predictions is the ESP model presented by Xapsos et. al. in 2000, which was previously discussed in section 3.1.3. The approach in this model is to use the principle of maximum entropy in order to generate a probability distribution of the proton flux [Xapsos et al., 2000], which can also be used for GCRs. For the short-term predictions, this model bases its analysis on fluxes measured in during an SPE in October 1989 on three different occassions: October 19th, October 22nd and October 24th. The distribution of proton flux can then be derived for a mean composition as well as a worst-case event.

(27)

3.2.3 Comparing the models

The input data for comparing the models for short-term predictions are the same as in the previous section 3.1.5, shown in table 4 for an orbit representing the moon. The resulting predicted flux is shown in figure 6 and figure 7 for the CREME 86 model, in figure 8 for the CREME 96 model and in figure 9 and figure 10 for the Xapsos model. For the CREME 86 model, figure 6 shows the predicted levels of of predicted flux for a mean composition, while figure 7 shows the same prediction for a worst-case composition. With level M5 and M6 as an exception, the mean composition and worst-case composition show the same predicted levels of solar flux for each scenario. Comparing the three different scenarios for predictions made by the CREME96-model, shown in figure 8, it can be seen that the prediction of a 5 minute peak gives a higher flux than the "worst day" and the

"worst week" scenario, where the lowest predicted flux is obtained for the worst week scenario.

When comparing the different predicted flux obtained by the ESP-model, shown in figure 9 and 10, the results show a similarity to the predicted flux given by the CREME-86 model. The mean value as well as the worst case scenario give the same predicted flux, the only difference for this model is for the different input data where the predictions are based on different events.

Figure 6: The level of solar proton flux at a trajectory following the moons orbit, derived with the CREME86 model. The plot shows the 4 different predicted levels of solar proton flux, for different values of M and mean-composition.

(28)

Figure 7: The level of solar proton flux at a trajectory following the moons orbit, derived with the CREME86 model. The plot shows the 4 different predicted levels of solar proton flux, for different values of M and worst-case composition.

(29)

Figure 8: The level of solar proton flux at a trajectory following the moons orbit, derived with the CREME96 model

(30)

Figure 9: The level of solar proton flux at a trajectory following the moons orbit, derived with the Xapsos et.

al. model for a mean case scenario.

(31)

Figure 10: The level of solar proton flux at a trajectory following the moons orbit, derived with the Xapsos et. al. model for a worst case scenario.

3.3. Real-time forecasts

The previously discussed models for predicting SPE’s rely on the probability of SPE occurrence, but for an ongoing mission this is not enough. To ensure the safety of the crew members of a manned spacecraft, it is crucial to have real-time forecasts of events instead of predictions based on probabilities. Space weather laboratories such as the Space weather prediction center (SWPC) monitors and forecasts the Earth’s space environment, as for example solar activity, and distributes this time-current information on their webpage [SWPC, 2016]. This information can then be used as input data in order to present real-time forecasts which make it possible to not only find the expected intensity profile of an ongoing event but also to get a warning of an upcoming event and thereafter ongoing forecasts until the event has ceased. In the case of a hazardous event, these forecasts give the crew of a manned spacecraft time to move to a more shielded area and making it possible to plan safer EVA’s.

(32)

3.3.1 UMASEP

This report will focus on discussing the UMASEP-model for forecasting SPEs. This model for real-time predictions is used by both NASA’s iSWA as well as ESA’s SEPsFLAREs space weather systems and a comparison between this model and other available forecasting models will be discussed at the end of this section. UMASEP was presented in 2011 by Marlon Núñez as a way of predicting solar particle events with an energy E>10MeV [Núñez, 2011]. Núñez explains in this report that systems designed to predict radiation events should do so both early as well as reliably.

It is important that they don’t miss any occuring events, but at the same time is it important that the rate of issued false warnings is at an acceptable level. This could otherwise affect and even delay ongoing space missions.

The UMASEP system is based on two models, one model for forecasting well-connected SPE and one for forecasting poorly-connected SPE. These two models are then combined in an additional model for analyzing the information. The system issues warnings for proton fluxes surpassing the SWPC SPE threshold of J(E>10MeV)=10 pfu, where 1 pfu = 1 proton cm⁻²sr⁻¹s⁻¹.

The well-connected SPE forecasting model tries to identify early signs of a well-connected event by identifying CME and solar flare processes associated with a forthcoming SPE. The model then empirically estimates the magnetic connectivity for these processes. The model analyzes soft X-ray and differential proton fluxes between 9-500 MeV and if the model determines that there is a magnetic connection, as well as an associated solar flare of C7-class or greater, it issues a warning.

The poorly connected SPE forecasting model tries to identify early signs of poorly connected events by comparing the differential proton flux to a historical poorly-connected SPE in order to derive similarities. At the beginning of a poorly-connected event, there are rising levels of protons, but the integral proton flux is not yet greater than the SWPC SPE threshold. The model can still recognize these rising levels as a poorly-connected event in progress and if so the model sends out a preliminary forecast.

The analysis module receives these forecasts from both forecasting models, and in the case of a preliminary forecast being issued, it can filter out the ones which are interpreted as erratic.

When analysing the forecasts from the well-connected SPE forecasting model, the analysis module retreives real-time solar data and can associate the event with a flare and its active region. The analysis module also predicts the expected intensity of the event during its first hours.

Figure 11: Showing the modules for the UMASEP system for forecasting events [Núñez, 2011], as well as the input and output for the system.

(33)

So how does the forecasting models actually work? Magnetically well-connected SPEs mean that after an originating solar activity, the solar particles follow the spiral paths in the interplanetary magnetic field (I MF) all the way from the Sun towards Earth, making it possible to detect these well-connected events at 1 AU several minutes or sometimes even hours after the originating event.

The aim of this model is to determine whether solar particles associated with any solar flare or CME are magnetically well-connected. It is important to analyze this as it shows that particles are escaping CMEs and solar flare processes and are arriving at 1 AU along the magnetic field lines.

To do so, the model looks at the correlation between the X-ray flux and proton flux data which is measured at 1 AU by all available GOES satellites. At every time-step t the model performs five correlation calculations for each satellite and then selects the one with the highest level of correlation for analyzing. If there is a high correlation between the X-ray and the proton flux, the model makes an assumption that there is also a magnetic connection occurring and issues a forecast of an event. Núñez shows in this report that prior to most SPEs during solar cycle 22 and 23 a correlation between the proton flux and the X-ray flux correlation occurs and it can often be seen that the proton flux occurs about 30 min after the X-ray flux. The magnetic connection is a necessary condition for predicting these kinds of events. If there is no correlation between the X-ray and proton flux the model does not issue any forecast, and it does not take into regard the intensity or heliolongitude of the observed flares and CMEs. The model is also able to issue preliminary forecasts. This is done by checking if the associated flare is a class C7 flare or higher which means that it will be equal or higher to the SWPC SPE threshold. This preliminary forecast is then sent to the analysis module, where the module calculates the expected intensity of the event and distributes this information.

To evaluate the performance of the forecaster for well-connected SPEs the Probability of Detection (POD) as well as the False Alarm Rate (FAR) are analyzed. These two terms can be written as

POD= ^A

(A+C) ⁽¹²⁾

and

FAR= ^B

(A+B) ⁽¹³⁾

where A is the number of correct forecast when a SPE was forecasted and then occured, B is the number of false alarms when a SPE was forecasted but did not occur and C is the number of missed events when a SPE did occur but was not predicted. The optimal configuration of the model is to have a high POD at the same time as having a low FAR.

For poorly-connected SPEs, the particles are accelerated by interplanetary CME-driven shocks. It is difficult to estimate the locations and evolutions of these solar activities which causes uncertainties regarding the particles propagation and acceleration. There is also no magnetic connection between the originating solar event and what is called the observer, in this case a GOES satellite. Therefor, this forecasting model does not work in the same way as the previously discussed one. This model instead makes the assumption that a gradual rise in several proton fluxes that does not seem to decrease is evidence that the satellite is beginning to be connected to the shock. This gradual rise could end up surpassing the treshold limit of 10 pfu, but it could also decrease. This makes the analysis hard, therefor this model uses a regression system to predict when the rise of proton fluxes would end up becoming a poorly-connected SPE and when it would end up decreasing.

I.e. the model analyses if the behaviour of the proton flux is similar to poorly-connected SPEs

(34)

happening in the past and from this makes an assumption based on similarities. The aim of this model is to predict the expected time that it takes for the proton flux to have an energy greater than the SWPC SPE threshold.

The model is also able to predict the first hours of the SPE. For every 5 minutes, the model makes a new prediction which produces a time series of the predicted event. If the final predicted proton flux surpasses the threshold level of 10 pfu at some future time, the model issues a preliminary forecast of a poorly-connected SPE. This forecast gives an indication of the time and expected intensity during the first 7 hours of the event. The forecast is then sent to the analysis model which either confirms them or filters them out.

The final forecast presented by the analysis model can be regarded as a combination of the preliminary forecasts issued from the two previous models. The module validates the consistency of the preliminary warnings and then either accepts them or filters them out. If the warning is accepted it compares all the information with solar activity data retrieved from the SWPC Solar and geophysical event reports containing data of recent solar events. By doing so, the model can then calculate the maximum intensity of the first 7 hours of the event. For the well-connected events the analysis module also identifies from which active region the SPE originated. The warning time for the SPE is the time interval between when the prediction is issued and the proton flux of the event is equal or greater than the SPWC threshold.

Figure 12: The graphical output for the well-connected SPE forecasting model. The colored area show the forecasted event, while the actual event is shown as a graph. [Núñez, 2011]

An example for the output of the UMASEP system can be found at [http://spaceweather.uma.es, 2016] which can also be seen in the following figures. Figure 12 shows two forecasts for well- connected SPEs that occurred November 8th, 1987 (left) and April 21th, 2002 (right). The forecasts are shown as a colored area in the graph and can be compared to the graph of the actual event.

(35)

While the first forecast was a success with a warning time of 3 hours 45 minutes according to Núñez, the latter only had a warning time of 35 minutes. Figure 13 shows two forecasts for poorly-connected SPEs that took place on December 7th, 2006 (top) and April 16th, 1990 (bottom).

The issued warning time for these events where 20 hours 5 minutes and 22 hours 5 minutes, respectively.

Figure 13: The graphical output for the poorly-connected SPE forecasting model. The colored area (in yellow) shows the forecasted event, while the actual event is shown as a graph. [Núñez, 2011]

For validation, the UMASEP system forecasts are compared to the SPEs of solar cycle 22 and 23.

This shows that the systems has a probability of detection of all well- and poorly connected events of 81%, and a false alarm rate of 34%. These values are considerably high, but are taken directly from the article [Núñez, 2011]. The system also has an average warning time of 5 hours 19 minutes for succesfully predicted events. If separated into well-connected and poorly-connected events the average warning time is 1 hour 5 minutes and 8 hours 28 minutes, respectively. UMASEP also has a maximum warning time of 24 hours for very gradual SPEs. As listed in table 6.

Table 6: Average warning time for different types of events.

Type of event Time

Succesfully predicted events avg. 5 h 10 min Well-connected events avg. 1 h 5 min Poorly-connected events avg. 8 h 28 min

Gradual events max 24 h

(36)

In 2015, Núñez presented a new version of the UMASEP system [Núñez, 2015], denoted as UMASEP-100. This new model makes real-time predictions of the time interval for which proton fluxes with E>100 MeV, associated with flares >M3.5, are expected to surpass the SWPC threshold.

The model is then able to estimate the first 3 hours of the event. UMASEP-100 works in a similar manner as the older system, denoted UMASEP-10, correlating the solar activity with the near-Earth proton flux data to make an assumption if there is a magnetic connectivity or not. Although the two systems are similar, there are still some small differences. One example is that they use a different approach for their correlations. The UMASEP-100 bases these correlations on proton fluxes measured for different proton channels at the GOES satellites compared to the older model.

For UMASEP-100, the ability to predict poorly-connected events is no longer available as it has been found that for proton fluxes with an energy E>100 MeV this prediction is not necessary in order to make a reliable forecast. Figure 14 shows the forecast results for the UMASEP-100 system for a SPE occuring in March 7th, 2012. The left image (bottom) shows that a well-connected event was identified after the X5.4 flare while the estimated proton flux for this event is shown above, this figure is borrowed from Núñez article and the explanation of the figure is borrowed from there as well. The right image (top) is a forecast issued several hours later which shows that the forecasted proton flux level highly correlates with the actual event. The warning time for this event was 45 minutes.

Figure 14: The graphical output for the UMASEP-100 forecasting model. The colored area (in yellow) shows the forecasted event, while the actual event is shown as a graph. [Núñez, 2011]

For the UMASEP-100 there is also a difference in statistical performance compared to the UMASEP- 10. When looking at the POD and the FAR, the probability of detection for UMASEP-100 is 81%

and the false alarm ratio is 30%, while the average warning time is 1 hour 6 minutes. The UMASEP-10 shows a POD of 89% and a FAR of 23% with an average warning time of 3 hours and 58 minutes. For analysing the statistical performance of both models, the forecasts of the

(37)

UMASEP-100 are compared with the forecasts of UMASEP-10 during a time interval between January 1994 to September 2013 [Núñez, 2015] which is presented in table 7.

Table 7: Statistical performance for UMASEP-10 and UMASEP-100.

System version POD FAR Avg. warning time UMASEP-10 89% 23% 3 h 58 min

UMASEP-100 81% 30% 1 h 6 min

Núñez concludes by explaining that the magnetic connection associated with Earth can be very different from the magnetic connection associated with an interplanetary spacecraft, and a hazardous SPE could be moving at relativistic speeds. It is therefore crucial to get enough warning time in order to take precautionary actions in order to protect humans and equipment on board a spacecraft from these events. An on-site forecast model could be the only system able to do this.

The spacecraft could be equipped with detectors for soft X-ray and proton fluxes from 10 to 500 MeV and the UMASEP system would be well fit to be used as an forecasting system for space missions to the Moon, asteroids or to Mars.

3.3.2 The Laurenza technique

There are of course other models available for forecasting SPEs. One example was presented in 2009, called the Laurenza technique [Laurenza et al., 2009]. This is described as a technique to provide short-term warnings of an upcoming SPE. The model is based on flare location, flare size and evidence of particle acceleration as well as the intensity of the soft X-ray and type III radio emissions.

3.3.3 PROTONS

Another example is a model presented in 2008 called PROTONS [Balch, 2008]. This model is based on the assumption that there is a relationship between the occurence of SPE and the intensity of solar flare emissions. Balch’s method is based on the peak flux of the soft X-ray, the time-integrated X-ray, the location of the flare on the heliosphere, and whether or not radio burst of type II and/or type IV occurs. It is important to mention is that PROTONS is merely used as a decision aid for human SPE forecasting experts, and does not produce forecasts on its own.

3.3.4 The proton prediction system

A method developed by Kahler et. al. in 2007, called the proton prediction system (PPS), predicts SPEs following solar flares with an energy E>5 MeV[Kahler et al., 2007]. The model bases its prediction on historical observed SPEs and their intensity-time profiles, as well as intensity of the peak and the duration of the events. By analysing the solar flare peak, time-integrated X-ray, radio fluxes and location of the solar flare it can predict peak proton intensities for energies E>10MeV and also predicts the onset and rise times of the event.

(38)

3.3.5 The Posner model

Another model presented in 2007 was developed by Posner. This is a short-term forecast model for predicting SPEs which used an electron-based technique [Posner, 2007]. This technique looks at the difference in transit-time for flare accelerated electrons and relativistic ions. If the location of a flare associated with SPEs can be determined, this short-term forecast is then possible due to a number of observed events. One of these events is that relativistic electrons always arrive at 1 AU before non-relativistic protons originating from solar activity. The forecast is also based on the fact that the intensity increase for both electrons and protons depends on the magnetic connection, as well as the fact that the existence of significant correlations shows that the intensity and increase of early electrons can be used to forecast the upcoming proton intensity.

3.3.6 The difference in forecasts

Most of the models, with PROTONS as an exception, do not analyze the signature of the acceleration at the shock wave (type II radio emissions). Instead they are based on solar flare-associated data which is an easier and more effective approach [Núñez, 2015]. All of the aforementioned models have the similarity that they are all able to predict poorly-connected events. Out of the five different presented models above, all of them use the SWPC SPE event list as their reference for calculating their levels of POD and FAR, except for the Posner model. Since the Posner model does not use this as a reference, only the remaining four models (UMASEP, PPS, PROTONS, Laurenza) will be compared by looking at their level of POD and FAR (se table 8). It is desired that POD is as large as possible, while FAR should be as low as possible. Since the values of POD and FAR for the different models are calculated for different intervals in time, this had to be taken into account when comparing the models.

Table 8: Comparing the statistical performance of models used for forecasting solar proton events

Forecaster POD FAR

1995-2005

UMASEP 83% 30%

Laurenza technique 63% 42%

1996-2004

UMASEP 80% 35%

PROTONS 57% 55%

1997-2001

UMASEP 78% 30%

PPS 40% (56%) 50%

1995-2005: The Laurenza technique was validated by using the data from 75 SPEs between 1995 to 2005. There was originally 93 occurences of SPEs during this time period, but some events were excluded with certain conditions. For example, events were excluded if the location of the associated flare was located on the backside of the sun and also if the flare was <M2. This resulted in only 68 events. There was also seven events added where strong proton enhancements occurred with strong flares, and were larger than the SPWC threshold which indicates an event. This resulted in a POD of 63% and a FAR of 42%. For the UMASEP-system, if abbreviated for the same

A study of SPEs and the consequences of radiation exposure

A study of SPEs and the consequences of radiation exposure

SIGNE BJÖRNHOLDT BÖLL

A study of SPEs and the

consequenses of radiation exposure

Author:

Signe Björnholdt Böll signebb@kth.se

Royal Institute of Technology (KTH)

Supervisors:

Christer Fuglesang Oscar Larsson

February 22, 2017

Contents

1. Introduction

2. Understanding solar particle events

2.1. The theory of origin

3. Prediction of events

3.1. Long-term prediction

∑

∑

3.2. Short-term prediction

3.3. Real-time forecasts